Otoacoustic emissions and pass/fail separation using artificial neural network

Abstract

Transiently evoked otoacoustic emissions (TEOAE) were proved to correlate with pure tone hearing thresholds. This relationship is often described with linear multivariate model. As it is known, TEOAE is highly nonlinear phenomena, thus nonlinear model should be a more accurate in relating TEOAE and hearing threshold. The aim of this study- to check if taking into account nonlinearity in relationship between mean hearing level and cross correlation values as TEAOE features can perform better than linear with equal feature weights in separation of hearing impaired and normal hearing subjects. Results show that the gain in specificity at a fixed sensitivity is small, although statistical hypothesis testing confirmed that this difference still is statistically significant.